[packagekit] running without a head via cron

Matthew Miller mattdm at mattdm.org
Mon Mar 10 13:39:11 PDT 2008


On Mon, Mar 10, 2008 at 04:24:12PM -0400, Robin Norwood wrote:
> I meant to respond to this earlier, but I agree with Sebastian that
> complicated scheduling and notification type logic has to reside in a
> monitoring/configuration management system, not in an individual
> application.  If you try to bring that logic into individual
> applications, you end up with an overly complicated system that never
> does exactly what some users want, and is configured differently from
> everything else.

Definitely agree about the scheduling -- that's what cron is *for*. 

It's the failure notification that I'm concerned about. If your application
is supposed to e-mail results somewhere, and the network is down, you end up
with frequency×downtime×number-of-machines e-mails when the network comes
back up. That's no good.

But I am inclined to agree that PackageKit isn't the place for it either.

Perhaps the best thing is for the pkcron application to produce no output
whatsoever except to log results (and dbus messages? I admit to not being
fully up to speed there) and have some way of telling something when it's
done, at which point a second program takes that logged output and sends a
report. This can of course be the standard system log reporter (logwatch or
epylog), but many people want to know about package updates immediately, not
the next morning (especially when action may be required).

I have more thoughts on this but haven't fully developed them yet. :)


-- 
Matthew Miller           mattdm at mattdm.org          <http://mattdm.org/>
Boston University Linux      ------>              <http://linux.bu.edu/>



More information about the PackageKit mailing list